Speedy is shockingly good for qwen3.8 FN
Summary
The author benchmarks Speedy against Avalon and Strata for running a Qwen3.8 model locally, finding Speedy dramatically faster: 4x faster reading, generation tok/s around 105-110, and 26-31% shorter total wall-clock time than Avalon.
Similar Articles
Qwen 3.8 27B is faster than expected
A user reports that the Qwen 3.8 27B model achieves 50-60 tokens per second on dual 5060 TI cards, showing unexpected speed improvements over previous versions like Qwen 3.6.
Strata with Qwen3.8 Flash Next UD-Q4_K_XL
A user reports strong single-request decode speeds running Qwen3.8-Flash-Next in Unsloth's UD-Q4_K_XL quant on Strata with an RTX PRO 6000, showing roughly 2x faster throughput than vLLM on the same GPU across prose, counting, coding, and reasoning tasks.
Qwen3.8-Flash-Next on 2x RTX 5090 with 504 tok/s decode: QEngine 1.3.1 decodes ~1.5-1.9x faster than the next-best engine I've tried (numbers, caveats, and what still bites)
A detailed benchmark shows QEngine 1.3.1 (compiled for RTX 5090s) is the fastest Qwen3.8-Flash-Next engine for single-stream decode, ~1.4-1.9x faster than Strata and ~3x faster than tuned vLLM, with a KV-in-RAM mode that dramatically improves warm returns to idle conversations.
Qwen3.8: 27b vs flash next. We all know the benchmarks, but at least to me, the reality is a different story
A developer compares self-hosted Qwen3.8-27B-GGUF (Q4_K_XL via llama.cpp) against cloud Qwen3.8 Flash Next, arguing that despite worse benchmarks, the dense 27B delivers better code quality, fewer rework iterations, more concise reviews, and far lower token consumption for their spec-to-PR workflow.
The Qwen 3.6 35B A3B hype is real!!!
The author benchmarks small local LLMs, highlighting Qwen 3.6 35B A3B for its superior ability to map academic code to research papers compared to models like Gemma 4 and Nemotron 3 Nano.